Transfer Learning-Based Classification of Kidney Stone From Computed Tomography Images

Kidney stone disease occurs when certain substances such as uric acid, oxalate, and calcium accumulate in high concentrations in the urine. This disease is a manageable condition with early diagnosis and appropriate treatment. Image processing-based systems hold great promise for the automatic evaluation and classification of kidney stones. In this study, we proposed an image processing-based model for the classification of kidney stone diseases using the contrast limited adaptive histogram equalization (CLAHE), Laplacian filter, and transfer learning model from computed tomography (CT) images. The CLAHE was applied for image enhancement, while the Laplacian filter was used to sharpen details and reduce blurriness in CT images. Transfer learning models, including AlexNet, Xception, VGG19, and ResNet101 were compared. The ResNet101 transfer learning model demonstrated the best performance, with 0.991 recall, 0.974 specificity, 0.966 precision, 0.978 f1-measure, 0.962 kappa statistic, 0.981 weighted-f1 score, and overall accuracy of 98.15%. This method provides a reliable and objective diagnostic tool, which could lead to more timely interventions.